| # 🛰️ ZETIC.ai — On-Device AI for Every Device | |
| **Build. Deploy. Run. Anywhere.** | |
| ZETIC.ai helps AI engineers deploy models on *any* mobile device — without cloud GPU servers. | |
| We transform your existing AI models into **NPU-optimized, on-device runtimes** in **under 6 hours** including from global device benchmark to runtime source code generation. | |
| --- | |
| ## 🚀 What We Do | |
| **ZETIC.MLange** — our core platform — enables **serverless AI** by: | |
| - **Automated Conversion**: Convert your PyTorch, ONNX, or TFLite model into a device-specific NPU library. | |
| - **Peak Performance**: Up to **60× faster** than GPU cloud inference, with zero accuracy loss. | |
| - **Broad Compatibility**: Supports Android, iOS, Linux; MediaTek, Qualcomm, Apple NPUs — more coming soon. | |
| - **End-to-End SDK**: From model optimization to app integration — no extra engineering required. | |
| --- | |
| ## 🛠 Key Features | |
| - **Zero GPU Costs** — Replace expensive GPU cloud servers with *free* NPU power in devices. | |
| - **Full Privacy & Security** — Data never leaves the device. | |
| - **Ultra-Low Latency** — Real-time AI experiences, even offline. | |
| - **Cross-Platform** — One model → All devices → Same performance. | |
| --- | |
| ## 📦 Example Use Cases | |
| - 🎙 **Speech Recognition (Whisper)** — Real-time, offline transcription on mobile. | |
| - 🦷 **Dental AI Diagnostics** — Instant tooth condition analysis via smartphone camera. | |
| - 🏌️ **Sports AI** — On-device golf swing analytics. | |
| - 🤖 **On-Device LLMs** — Chat & reasoning models running entirely offline. | |
| --- | |
| ## 📊 Benchmarks | |
| | Device | Task | Cloud GPU | On-Device NPU | Speedup | | |
| |--------|------|-----------|---------------|---------| | |
| | iPhone 16 Pro | Whisper-Small | 1.2s | 0.07s | **×17** | | |
| | Galaxy S24 Ultra | LLaMA-3-8B | 2.4s/token | 0.09s/token | **×26** | | |
| [🔗 See more benchmarks »](https://mlange.zetic.ai) | |
| ### YOLOv8n — NPU Latency (ms) | |
| | Device | Manufacturer | CPU | GPU | CPU/GPU | NPU | | |
| |--------|--------------|-----|-----|---------|-----| | |
| | Apple iPhone 16 | Apple | 126.27 | - | 8.98 | **2.03** | | |
| | Apple iPhone 16 Pro | Apple | 122.23 | - | 7.54 | **1.69** | | |
| | Samsung Galaxy S24+ | Qualcomm | 69.79 | 24.38 | 618.05 | **3.85** | | |
| | Samsung Galaxy Tab S9 | Qualcomm | 107.78 | 30.39 | 344.42 | **5.21** | | |
| | Samsung Galaxy S22 Ultra 5G | Qualcomm | 103.40 | 39.73 | 100.34 | **7.41** | | |
| --- | |
| ### Whisper-tiny-encoder — NPU Latency (ms) | |
| | Device | Manufacturer | CPU | GPU | CPU/GPU | NPU | | |
| |--------|--------------|-----|-----|---------|-----| | |
| | Apple iPhone 16 | Apple | 552.13 | - | 44.49 | **19.01** | | |
| | Apple iPhone 15 Pro | Apple | 527.78 | - | 43.13 | **19.40** | | |
| } Samsung Galaxy S23 | Qualcomm | 290.62ms | 169.82ms | 2,795.18ms | **86.88** | | |
| | Samsung Galaxy S24+ | Qualcomm | 278.78 | 133.48 | 2619.56 | **106.44** | | |
| | Samsung Galaxy S23 Ultra | Qualcomm | 308.82 | 170.08 | 2688.97 | **68.34** | | |
| - **You can get runtime source code and benchmark report of your model with [ZETIC.MLange](https://mlange.zetic.ai)** | |
| --- | |
| ## 👨🏻💻 Plug-and-play To Your App | |
| - The runtime SDK is also provided for your AI model with ZETIC.MLange | |
| - **iOS Integration** (Swift) | |
| ``` swift | |
| // import | |
| import ZeticMLange | |
| // ... | |
| // (1) Load Zetic MLange model | |
| let model = try ZeticMLangeModel("MLANGE_PROJECT_API_KEY") | |
| // (2) Run model after preparing model inputs | |
| let inputs: [Data] = [] // Prepare your inputs | |
| try model.run(inputs) | |
| // (3) Get output data array | |
| let outputs = model.getOutputDataArray() | |
| ``` | |
| - **Android Integration** (Kotlin, Java) | |
| ``` kotlin | |
| // import | |
| import com.zeticai.mlange.core.model.Target | |
| import com.zeticai.mlange.core.model.ZeticMLangeModel | |
| // ... | |
| // (1) Load Zetic MLange model | |
| val model = ZeticMLangeModel(this, "MLANGE_PROJECT_API_KEY") | |
| // (2) Run model after preparing model inputs | |
| val inputs: Array<ByteBuffer> = // Prepare your inputs | |
| model.run(inputs) | |
| // (3) Get output buffers of the model | |
| val outputs = model.outputBuffers | |
| ``` | |
| ## 📥 Try It Now | |
| - **MLange Dashboard**: [https://mlange.zetic.ai](https://mlange.zetic.ai) | |
| - **Demo Apps**: [App Store](https://apps.apple.com/app/zeticapp/id6739862746) / [Google Play](https://play.google.com/store/apps/details?id=com.zeticai.zeticapp) | |
| --- | |
| ## 🧭 Supported Targets | |
| - **OS**: Android, iOS, Linux | |
| - **NPUs**: MediaTek, Qualcomm, Apple (more coming) | |
| - **Frameworks In**: PyTorch, ONNX, TFLite | |
| - **Artifacts Out**: NPU-optimized runtime libraries + SDK bindings (Kotlin, Java, Swift, Flutter, React Native) | |
| ## 📬 Contact Us | |
| - **Website**: [https://zetic.ai](https://zetic.ai) | |
| - **Email**: [email protected] | |
| - **LinkedIn**: [linkedin.com/company/zetic-ai](https://linkedin.com/company/zetic-ai) | |
| --- | |
| **ZETIC.ai** — AI for All, Anytime, Anywhere. | |
| Run your AI where it matters: **on the device.** | |